Redesigning the Snackbot GUI

UX HCI, on December 29th, 2010 go back to index

The Redesign of the Snackbot GUI was the assignment for the lab part of Methods, at Carnegie Mellon University (CMU). For this reason some of the paths followed or decisions taken were made to fit the course requirements.

What is snackbot?

From www.snackbot.org

Snackbot is a mobile robot, about the size of a very small human, that rolls around on wheels, and will be delivering snacks to students, faculty, and office workers at Carnegie Mellon University. A Snackbot research team of faculty, graduate students, and undergraduates from more than three departments at Carnegie Mellon created Snackbot as a platform for studying human-robot interaction in the real world.

Snackbot is meant as an ongoing platform for research. The Snackbot will support research into robust autonomous operation in office environments. Our efforts range from multi-sensor fusion algorithms for perception, reasoning about dynamic spaces,communicating with people through verbal and non-verbal mechanisms, and planning with incomplete information.

In order for the snackbot to be used as a research platform or even just to deliver cookies, it needs humans to program it’s behaviour. This way, the software developers created three interfaces for researchers to create head animations, mouth animations and dialog trees. All of these interface’s outputs were stackable. For instance, a mouth animation would be created and saved. This file would be fed into the head animation GUI so that it is merged with it. This file was then saved to a single file that would be finally fed into the dialog tree GUI.

The task

Our task was to design a software workflow that assists researchers create human-like expressions and animations for Snackbot. The proposed workflow will improve the productivity and efficiency of researchers creating animations and expressions for Snackbot.

Gathering data

Before we started to gather data, we needed a focus. To do this the team brainstormed ideas, issues and themes related to the project. This list was organized into an affinity diagram to structure the data into chunks. This process allowed us to share our knowledge with each other and have a group understanding of what the project is about and what we should focus on.

Using the affinity diagram, we made a focus setting meeting to define our focie. We came up with two.

1. Human-Robot Interaction – Finding out how researchers run experiments with Snackbot. Since the users will be using the robot for reaseach of human robot interaction it makes sense to understand that part of their work.

2. Interface – Discussing the elements missing on the current interface and finding the most natural way to script animation on the robot. Again with the intent of supporting the researchers since they will need to animate the robot, ideally without any help from the technical people.

From these two focie we conducted contextual inquiries on experts in Animation and current researchers of snackbot. The Animators were students from ETC that showed us how 3D software was used to programe a sequence of movements and actions. The researchers showed us their whole process and methods to create a snackbot script.

Modeling

We consolidated all of this information into work, cultural, sequence and artifact models. The consolidation of our findings from the researchers was separated from the animations because of their radically different domains and focuses.

We noticed instances of role strain and role sharing among the Snackbot researchers. The GUI programmer often collaborated closely with the animator and compiler, and in some situations performed all three roles. Multiple people generated Snackbot expressions and all of these would be sent to a single person who would combine and input the assets into Snackbot.

Using the cultural model we understood the amount of pressure presented to the Research Assistants (RAs). Most of them are appeals for a better tool that is user friendly, does not force the user to remember software constraints and provides accurate feedback from the robot.

The sequence models gave us critical information about the process that the researchers undertake in order to create a snackbot script. Clearly, the digital creation of the script presented major problems.

From the Animations at ETC we learned much concerning the software for animation. We were particularly interested in the notion of timeline and how the experts use it to create and preview animations. We took these timelines as artifacts for our reference.

Design Ideas

All of this information was good fuel to create new designs. To captivate the groups imagination had another brainstorm session to organize our design ideas and start envision.

Five categories were predominant in our affinity diagram: Human-Like Robot Behavior, Colaboration and Reuse, Feedback, Robot Restaints and a single unified tool.

Making Snackbot more human-like.

One fact we observed was the desire by both researchers and faculty to make Snackbot more human-like. Future versions of the Snackbot animation application should make it easier for users to create human-like expressions and head movements.

Affording feedback.

We noticed in our contextual interviews the unresponsiveness of the current Snackbot animation interface and its inability to effectively convey important aspects of the Snackbot animation workflow to the user. The Snackbot animation application should show the most important features forefront in the interface and provide responsive feedback for all interactors in the interface.

Indicating Snackbot constraints.

Our interviews with the Snackbot researchers showed that there were many hardware constraints with Snackbot, causing limitations on the expressions possible with Snackbot. Many of these limitations were initially unknown to the researchers and were discovered through experience. To prevent a repeat of this situation, the interface should warn users of expressions that may be impossible to perform with the current Snackbot hardware.

Supporting collaboration and reuse.

Our research has also shown that different people collaborate in overlapping areas of the Snackbot research. Additionally, many of the research assistants are Carnegie Mellon undergraduates who only can commit several years of work to the Snackbot project. As a result, the interface should make it easy for users to collaborate on a Snackbot project, and facilitate reuse of expressions and assets that have been previously created.

Merging interfaces

We have also observed that the researchers begin their work by first creating a script. The animations and expressions will then be tailored to specific points in the script to express an emotion. Due to this tightly integrated process, it would be unnatural to separate the work into different interfaces as it is currently being done. The new design should integrate and support this mode of work by building the assets from the script and integrating the script, mouth expression and head animation into one.

Prototyping and Testing

With those design ideas in mind we fashioned a prototype. Our first iteration consisted of an interface with two parts. The dialog creation part and a “zoomed” view for each dialog for speech, mouth expression and head movement defenitions.

We then conducted think aloud usability studies on 4 novice users with no prior experience with animation software. The usability studies were conducted with paper prototypes using “wizard of oz” manipulation of the prototype, following Think Aloud principles. The users are asked to perform 2 simple tasks:

1. Create two nodes and connect them with a transition with a simple speech as the transition type.

2. Go to the editor. Create smile and pan animations from the presets library and modify one of the frames.

Most of the users are able to complete the tasks successfully. Some usability issues with the prototype are identified through the study:

• Users are unable to identify the functionalities of some buttons based on the icons alone. The grouping of unrelated functionality of the buttons confused some users.

• The prototype uses some ambiguous phrases in the interface that the user is unable to understand.

• Some of the interface interactions use drag and drop which is not easily discoverable by users. The mental model in the interface is not consistent as some parts use button clicks and others drag and drop.

• Some of the users are not aware of the concept of key frames and how to manipulate them.